Parallel computing with ODEs

Hi,

I am having a problem: I need to solve a stiff system composed by more than 1500 ordinary differential equations (using the ode15s function). The principal problem is computational time and I am not familiar in optimizing it. We have a dual core processor and I suppose that without any special order Matlab is not exploiting all its power.

First question, is the parallel computing toolbox ok for this task?

Second question, having 1500 interconnected differential equations (the first one affects all the others, and the n-th one is affected by all the previous ones), is it possible to reduce the computational time using parallel computing? In my head, equations should be splitted in 2 groups, which are divided in the two cores, but these two cores should communicate between them because equations in the second group will need information contained in the equations in the first group.

Thank you for your help.

Answer 1

[Rui]

I doubt parallel computing with dual core will help. But you still have two options, first, see if your company or school has a grid system, I do not how to call it exactly, but basically there you get multiple cores, e.g., 32 cores; second, do not use default matlab ode solvers, you can try this one http://bmi.bmt.tue.nl/sysbio/software/CVode.html, which generates compiled mex files, but you need to figure out how to set it up on your machine. You can combined the two options to save even more time.

[Edoardo Melloni]

First of all, thank you for your answer. As I wrote in the question, I do not have much experience in computer components, I'm just a final user. I know we have a cluster with 96 processors or cores, I don't even figure out if I need more processors or more cores. Thanks for the link, I'm gonna see it and try to figure out something from it!